• 한국융합학회논문지
• /
• 제12권11호
• /
• pp.109-117
• /
• 2021

최근 인간과 컴퓨터의 상호작용(HCI)을 위한 수단으로 음성기반 인터페이스의 사용률이 높아지고 있다. 이에 음성인식 결과에 오류를 교정하기 위한 후처리기에 대한 관심 또한 높아지고 있다. 그러나 sequence to sequence(S2S)기반의 음성인식 후처리기를 제작하기 위해서는 데이터 구축을 위해 human-labor가 많이 소요된다. 최근 기존의 구축 방법론의 한계를 완화하기 위하여 음성인식 후처리기를 위한 새로운 데이터 구축 방법론인 Back TranScription(BTS)이 제안되었다. BTS란 TTS와 STT 기술을 결합하여 pseudo parallel corpus를 생성하는 기술을 의미한다. 해당 방법론은 전사자(phonetic transcriptor)의 역할을 없애고 방대한 양의 학습 데이터를 자동으로 생성할 수 있기에 데이터 구축에 있어서 시간과 비용을 단축할 수 있다. 본 논문은 기존의 BTS 연구를 확장하여 어떠한 기준 없이 데이터를 구축하는 것보다 어투와 도메인을 고려하여 데이터 구축을 해야함을 실험을 통해 검증을 진행하였다.

• 대한음성학회지:말소리
• /
• 제43호
• /
• pp.57-72
• /
• 2002

The technology of speech recognition has a wide field of application. The range of such technology is spreading into mobile computing having the large amount of movement for communication equipments at the present time. Particularly, recognition in internet environment is rapidly moving into mobile environment. Because of these environments, users want the faster speed of data transmission and the lighter portable equipment for data access. That is PDA(Personal Digital Assistant). Therefore, we designed a triphone-based text independent vocabulary recognizer for the implementation of speech control in this paper. The text independent vocabulary recognizer is based on the state .joint algorithm with decision trees

• 대한음성학회지:말소리
• /
• 제43호
• /
• pp.127-136
• /
• 2002

The speaker adaptation technique is generally used to reduce the speaker difference in speech recognition. In this work, we focus on the features fitted to a linear regression-based speaker adaptation. These are obtained by feature transformation based on independent component analysis (ICA), and the transformation matrix is learned from a speaker independent training data. When the amount of data is small, however, it is necessary to adjust the ICA-based transformation matrix estimated from a new speaker utterance. To cope with this problem, we propose a smoothing method: through a linear interpolation between the speaker-independent (SI) feature transformation matrix and the speaker-dependent (SD) feature transformation matrix. We observed that the proposed technique is effective to adaptation performance.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2002년도 11월 학술대회지
• /
• pp.119-122
• /
• 2002

This paper focuses on building up a database of commercial stocks using XML syntax and looks into a way of building up a system with the combination of XML and XSLT that provides connectivity to client-server databases through vocal means. The use of XSLT has several advantages. Most importantly, it can transform a type of data into different formats. A vocal interface minimizes some space and time limits imposed on users outside premises when they need an instant connection to their database. In this fashion, the users can check information on stock lists without being pressurized by certain limits. PC, PDAs and cellular phones are some examples of mobile connection. The use of VoiceXML creates vocal applications. In VoiceXML servies, users can gain immediate access to data upon the input of their voices and the DTMF signals of the telephone.

• ETRI Journal
• /
• 제32권5호
• /
• pp.810-818
• /
• 2010

This paper presents a new framework for integrating untranscribed spoken content into the acoustic training of an automatic speech recognition system. Untranscribed spoken content plays a very important role for under-resourced languages because the production of manually transcribed speech databases still represents a very expensive and time-consuming task. We proposed two new methods as part of the training framework. The first method focuses on combining initial acoustic models using a data-driven metric. The second method proposes an improved acoustic training procedure based on unsupervised transcriptions, in which word endings were modified by broad phonetic classes. The training framework was applied to baseline acoustic models using untranscribed spoken content from parliamentary debates. We include three types of acoustic models in the evaluation: baseline, reference content, and framework content models. The best overall result of 18.02% word error rate was achieved with the third type. This result demonstrates statistically significant improvement over the baseline and reference acoustic models.

• 대한음성학회지:말소리
• /
• 제49호
• /
• pp.135-144
• /
• 2004

In this paper, we make a study of classification normal from abnormal - normal, asthma through analysis of thoracic sound to take use thoracic sound detection system. Thoracic sound detection system has a function to store thoracic sound and analyze the data. The wave shape of thoracic sound is similar to noise and is systematically generated by inhalation and exhalation breathing, therefore, in this paper, to classify asthma sound in thoracic sound, we could discriminate between normal and abnormal case using level crossing rate(LCR) and spectrogram energy rate.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2003년도 10월 학술대회지
• /
• pp.141-143
• /
• 2003

Speech recognition performance depends on various factors. One of the factors is the characteristic and established distance of a microphone which is used when speech data is collected. Thus, in the present experiment speech databases for tests are created through the type and established distance of a microphone. Then, acoustic models are built based on these databases, and each of the acoustic models is assessed by the data to determine recognition performance depending on various microphones and established microphone distances.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2003년도 10월 학술대회지
• /
• pp.69-72
• /
• 2003

In this paper, we investigate various factors that are relevant to design and implementation of an integrated management system for various speech corpora. The purpose of this paper is to manage an integrated management system for various kinds of speech corpora necessary for speech research and speech corpora consrtructed in different data formats. In addition, ways are considered to allow users to search with effect for speech corpora that meet various conditions which they want, and to allow them to add with ease corpora that are constructed newly. In order to achieve this goal, we design a global schema for an integrated management of new additional information without changing old speech corpora, and construct a web-based integrated management system based on the scheme that can be accessed without any temporal and spatial restrictions. And we show the steps by which these can be implemented, and describe related future study topics, examining the system.

• 대한음성학회지:말소리
• /
• 제53호
• /
• pp.93-102
• /
• 2005

The performance of a speech recognition system is generally degraded in telephone environment because of distortions caused by background noise and various channel characteristics. In this paper, data-driven temporal filters are investigated to improve the performance of a specific recognition task such as telephone speech. Three different temporal filtering methods are presented with recognition results for Korean connected-digit telephone speech. Filter coefficients are derived from the cepstral domain feature vectors using the principal component analysis. According to experimental results, the proposed temporal filtering method has shown slightly better performance than the previous ones.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2007년도 한국음성과학회 공동학술대회 발표논문집
• /
• pp.91-94
• /
• 2007

This paper aims to analyze the pronunciation of Korean Point-of-Interest (POI) data, which consist of 224 sound files, from the phonological point of view, adapting the notion of prosodic word within the framework of Intonational Phonology. Each POI word is broken down into prosodic words, which are defined as the minimal sequence of segments which can be produced as one Accentual Phrase (AP). Then the pronunciation of the POI word considering its prosodic words are analyzed. The results show that: in most cases, a prosodic word is realized as one AP; that, in some cases, two prosodic words are pronounced as one AP: and that no cases are found where 3 prosodic words are realized as one AP.